📰 Key Takeaways

Bitcoin security researcher Rob Hamilton (CEO of AnchorWatch) posted on X that OpenAI restricted his access to Trust & Cyber features, forcing him to switch back to Chinese open-source AI models to continue his Bitcoin infrastructure security research. He’d only started integrating OpenAI’s Trust & Cyber capabilities into Bitcoin Red Team’s research work last Saturday — by the next morning, his access had already been restricted. Hamilton described it as “genuinely painful as a patriotic American,” but said he’s now decided to go back to using Chinese open-source models to keep the research going. Bitcoin Red Team is a group of volunteers using AI tools combined with manual review to scan hundreds of Bitcoin-related open-source repositories for vulnerabilities — as of Saturday, they’d found a cumulative 1,288 critical and high-severity vulnerabilities. This scanning push accelerated after the Coldcard hardware wallet hack, which resulted in losses of over $100 million in bitcoin. Hamilton says he can no longer continue his investigation, and has no way to confirm whether the relevant code patches are sufficient, or whether other undiscovered issues remain. He criticized the move, saying “black hat hackers aren’t affected by these restrictions at all — it’s the white hats who get blocked. Our policy has hit a local minimum,” and stressed that “people who don’t play by the rules get unlimited access to AI intelligence, while the people working to reduce harm get left out in the cold.” Last month, crypto industry executives told Cointelegraph that many major industry players are still waiting for access to powerful new AI models to strengthen their code defenses — currently, only a handful have been granted access.

💬 JudyAI Lab’s Take

As an observer deeply embedded in the AI space, JudyAI Lab noticed a case worth documenting today: Bitcoin security researcher Rob Hamilton was forced to switch to Chinese open-source AI models to continue his vulnerability-scanning research after OpenAI restricted his access to Trust & Cyber features — highlighting how AI access permissions are becoming a new variable in security defense.

Bitcoin Red Team has so far used AI tools combined with manual review to uncover 1,288 critical and high-severity vulnerabilities, with this push accelerating after the Coldcard hardware wallet hack that resulted in losses of over $100 million. Hamilton’s situation points to a structural problem worth thinking about for AI builders: when access to powerful models gets tightened or selectively granted, it’s usually the people playing by the rules — the white hats publicly disclosing vulnerabilities — who bear the cost, while attackers who don’t follow the rules aren’t bound by these restrictions at all. It’s a reminder that tool availability itself is a security variable, not just a matter of performance or cost.

For AI builders, it’s worth tracking not just how capable a model is, but also whether the access policies for the AI tools you depend on are stable — and whether your critical research work needs a backup plan.

📅 Original Article Info

🔗 Further Reading